installation of CP2K with support of CUDA

102 views
Skip to first unread message

Nikhil Maroli

unread,
Aug 7, 2019, 7:22:31 AM8/7/19
to cp2k

Dear all,

Im new here with CP2K.  My system has 3 GPU card (GTC 1070) and 2 x 12 core processor.
All the packages are installed and I have gone for   

./install_cp2k_toolchain.sh --install-all command and all the files are generated and kept in arch folder.


The end results are given below

Done!
Now copy:
  cp /home/user2/cp2k-6.1.0/tools/toolchain/install/arch/* to the cp2k/arch/ directory
To use the installed tools and libraries and cp2k version
compiled with it you will first need to execute at the prompt:
  source /home/user2/cp2k-6.1.0/tools/toolchain/install/setup
To build CP2K you should change directory:
  cd cp2k/makefiles/
  make -j 48 ARCH=local VERSION="sopt sdbg ssmp popt pdbg psmp"

arch files for GPU enabled CUDA versions are named "local_cuda.*"
arch files for valgrind versions are named "local_valgrind.*"
arch files for coverage versions are named "local_coverage.*"


I would like to run CP2K with my GPU+CPU, kindly instruct me what step i need to take further to install GPU version of the software.


I have tried
  make -j 48 ARCH=local_cuda VERSION="sopt sdbg ssmp popt pdbg psmp"

results are here

/home/user2/cp2k-6.1.0/makefiles/Makefile:129: recipe for target 'all' failed
make[2]: *** [all] Error 2
/home/user2/cp2k-6.1.0/makefiles/Makefile:118: recipe for target 'pdbg' failed
make[1]: *** [pdbg] Error 2
/home/user2/cp2k-6.1.0/makefiles/Makefile:129: recipe for target 'all' failed
make[2]: *** [all] Error 2
/home/user2/cp2k-6.1.0/makefiles/Makefile:118: recipe for target 'sdbg' failed
make[1]: *** [sdbg] Error 2
Makefile:113: recipe for target 'all' failed
make: *** [all] Error 2


Jeronimo Abello

unread,
Jan 8, 2020, 7:28:10 PM1/8/20
to cp2k
HI,

I am trying to download CP2K on Ubuntu and I get the same error. DId you found a way to solve it ?

Thanks.

lizzy bai

unread,
Jan 8, 2020, 9:15:16 PM1/8/20
to cp...@googlegroups.com
Hi.

I thought  you need modify some information in the loca_cuda file.  
"NVFLAGS     = -arch sm_61 $(DFLAGS)" 
In my case I change the number to "61",since the compute capability is 61 for Titan XP.
So you need check the compute capability for you GPU (GTC 1070) then modify that number.

Hope this helps

Best wishes

Liyi


Jeronimo Abello <jero...@gmail.com> 于2020年1月9日周四 上午9:28写道:
--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/b58d0d0e-6e13-4e0d-8796-e91b01cd725a%40googlegroups.com.
Reply all
Reply to author
Forward
0 new messages